Head to head: AuraFlow vs Cosmos 3 Super
AuraFlow vs Cosmos 3 Super
By Ryan Merket · Published
Cosmos 3 Super looked more reliable on prompt obedience, but the aggregate margin isn’t trustworthy enough to call this matchup for either side. In practice, AuraFlow and Cosmos 3 Super finish effectively even here, with each model exposing a different kind of weakness.
This is a tie, full stop. Yes, Cosmos 3 Super posts the higher aggregate score, 66.6 to AuraFlow’s 48.7, but the statistical read says **too close to call** — just **50% confidence** that either model is genuinely better. That is not a verdict; that is a coin flip wearing a spreadsheet. The shape of the matchup is still clear. Cosmos 3 Super was the steadier rule-follower on the tasks that punish sloppiness: it handled the **typography kiosk poster** far better, repeatedly rendering **"NOVA WHARF 7"** and **"FERRY OPENS 05:40"** legibly while AuraFlow mangled the text. It was also stronger on **exact counting**, **perspective and scale**, and most of the scene-construction briefs like the **teal-coral alley** and **car-free market avenue**, where AuraFlow had a bad habit of breaking core constraints — especially by inserting cars where the prompt explicitly banned them. AuraFlow’s case rests less on compliance and more on style. It was the more convincing model on the **named art style** task when it locked into a true ukiyo-e woodblock feel: flatter color planes, bolder outlines, more convincing stylized foam. And on **neon noodle bar night**, the judges effectively canceled each other out. One pass preferred Cosmos 3 Super’s moodier, more cinematic intersection read; the other preferred AuraFlow’s lower eye-level diagonal setup and stronger corner composition. Net result: dead even. The messiest category was **attribute binding**, where neither model exactly covered itself in glory. Cosmos 3 Super usually preserved the scene logic better, but there were runs where it invented the wrong blue object or put the duck on the wrong block; AuraFlow often looked nicer while still breaking the actual object relationships. That split is the whole story of this matchup: Cosmos 3 Super was often more obedient, AuraFlow sometimes more artistically persuasive, and neither one was consistent enough to earn a clean editorial endorsement. **Final call: TOO CLOSE TO CALL. Cosmos 3 Super may have looked like the safer prompt-follower, but at 50% confidence this matchup is effectively even, and declaring a winner would be fake precision.**
Typography kiosk poster
A rainy urban sidewalk at dusk, photographed straight-on in a crisp editorial style, centered on a glass-sided transit kiosk displaying a clean poster with the exact headline text: "NOVA WHARF 7" and, directly below in smaller type, "FERRY OPENS 05:40"; the lettering must be perfectly legible, correctly spelled, and easy to read from edge to edge, with realistic typography, subtle reflections on wet pavement, pedestrians softly blurred in the background, cool blue ambient light and a warm ad-box glow, 16:9.


Model B adheres closely to the prompt with a centered transit kiosk, rainy dusk street, strong editorial composition, and perfectly legible text reading exactly "NOVA WHARF 7" and "FERRY OPENS 05:40." Model A has appealing mood and reflections, but the typography is badly corrupted and misspelled, which is a major failure for a typography-focused poster prompt. (Second judge pass, order swapped — scores are the average of both: Model B closely matches the prompt with a straight-on editorial view of a glass-sided transit kiosk, strong rainy dusk atmosphere, and fully legible text reading exactly "NOVA WHARF 7" and "FERRY OPENS 05:40." Model A has appealing mood and reflections, but it fails the typography requirement with heavily corrupted text, the wrong wording, and a less accurate kiosk/poster presentation.)
Neon noodle bar night
A cinematic urban street corner at neon night, painted in a highly detailed digital concept-art style, featuring a tiny noodle bar under a translucent orange awning, steam rising from sidewalk pots, reflective puddles, tangled overhead cables, and two patrons sharing a late meal at a metal table; the scene should be lit primarily by saturated magenta and cyan shop signs with deep surrounding shadows, creating a moody, electric atmosphere, low eye-level composition looking diagonally across the intersection, 16:9.


Model B adheres more closely to the prompt with a tiny noodle bar under a translucent orange awning, tangled overhead cables, steam, reflective puddles, and a diagonal-across-intersection urban composition that feels more cinematic and moody. Model A is striking and polished, but its framing is more centered and storefront-focused, with less of the requested low eye-level diagonal intersection view and more obvious text artifacts. (Second judge pass, order swapped — scores are the average of both: Model A matches more of the prompt’s key specifics: a low eye-level diagonal corner view, saturated magenta/cyan signage, translucent orange awning, visible steam pots, reflective puddles, and two patrons at a metal table. Model B is moody and polished, but its higher vantage point and less direct street-corner composition make it slightly less aligned with the requested cinematic setup.)
Exact counting
A clean overhead flat lay on white marble of EXACTLY seven identical espresso cups arranged in a single straight row, each clearly separate and individually countable, even soft lighting, 16:9.


Model B adheres much more closely to the prompt by presenting a clean overhead flat lay on white marble with cups arranged in a single straight row, though it shows only five cups instead of exactly seven. Model A is aesthetically strong and technically polished, but it fails the core counting and arrangement requirements by showing many more than seven cups in a grid rather than one row. (Second judge pass, order swapped — scores are the average of both: Model B is much closer to the prompt, presenting an overhead flat lay on white marble with cups in a single straight row, but it only shows five cups instead of exactly seven. Model A is aesthetically polished, yet it badly fails the counting and arrangement requirements by showing many more than seven cups in a grid rather than one straight row.)
Teal-coral palette alley
An isometric 3D render of a narrow urban service alley transformed into a stylish micro-courtyard cafe, viewed from above at a three-quarter angle, with stacked fire escapes, tiny metal tables, potted succulents, and a parked delivery scooter; restrict the entire image strictly to the named palette "Lagoon Signal" consisting only of deep teal, seafoam, muted coral, pale sand, and charcoal, with no other colors anywhere in the scene, clean geometric forms, soft matte materials, and even overcast daylight, 16:9.


Model B adheres more closely to the prompt with a convincing narrow urban alley turned micro-courtyard cafe, clear stacked fire escapes, tables, succulents, scooter, and a cleaner isometric three-quarter composition. Model A has appealing stylization but is weakened by prominent extraneous text, a less believable alley context, and weaker compliance with the strict scene framing despite staying closer to the requested palette. (Second judge pass, order swapped — scores are the average of both: Model B adheres more closely to the requested narrow urban alley turned micro-courtyard cafe, with the correct isometric-above viewpoint, stacked fire escapes, tables, succulents, scooter, and a palette that stays much closer to the specified teal-coral-sand-charcoal range. Model A has appealing stylization, but it breaks the prompt with poster-like text elements, a less convincing alley context, harsher lighting, and more noticeable palette and rendering deviations.)
Named art style
A ukiyo-e woodblock print of a fishing boat riding a large cresting wave at dawn, faithful to the flat color planes, bold outlines, and stylized foam of the tradition, muted indigo and cream palette.


Model A strongly evokes ukiyo-e and has striking stylized foam and bold outlines, but it leans very closely on a specific iconic composition rather than simply depicting the prompt. Model B better matches the requested fishing boat riding a large cresting wave at dawn with a muted indigo-and-cream palette, while maintaining a cohesive woodblock-print look and balanced composition. (Second judge pass, order swapped — scores are the average of both: Model A more convincingly matches the ukiyo-e woodblock tradition through flatter color planes, stronger stylized foam, and a more iconic composition, while also presenting a more dynamic large cresting wave at dawn. Model B is attractive and on-prompt, but it feels more like a modern illustration influenced by ukiyo-e rather than as faithful to the named style.)
Car-free market avenue
A lively weekend street market on Brindle Row in a dense city neighborhood, illustrated as a detailed isometric 3D render, with fruit stalls, striped awnings, folding chairs, chalkboard menus, cyclists walking their bikes, kids chasing soap bubbles, and a tram track running down the middle — but the street must contain absolutely no cars, vans, taxis, buses, or motorcycles anywhere; composition should clearly show the full avenue from corner to corner, bright spring daylight, 16:9.


Model B better matches the prompt’s car-free market avenue with a clear full-street composition, lively market details, cyclists, kids, bubbles, and a tram track without cars. Model A has strong detail but violates the prompt heavily by placing multiple cars on the avenue and also feels less aligned with the requested full corner-to-corner market scene. (Second judge pass, order swapped — scores are the average of both: Model B closely matches the prompt with a car-free market avenue, visible tram tracks and tram, striped awnings, fruit stalls, folding chairs, chalkboard menus, cyclists walking or riding alongside the market, and children chasing bubbles in a bright detailed scene. Model A violates the core requirement by showing multiple cars and motorcycles, lacks the tram track, and includes garbled text, despite a decent full-avenue composition.)
Attribute binding
A photorealistic tabletop scene: a matte green cube to the LEFT of a glossy red sphere, a small blue cylinder BEHIND both, and a single yellow rubber duck sitting ON TOP of the cube. Colors must stay on the correct objects. Soft studio light, 16:9.


Model B follows the prompt much more accurately: the matte green cube is to the left of the glossy red sphere, the blue cylinder is behind them, and the yellow rubber duck sits on top of the cube. Model A is visually polished, but it breaks the key attribute bindings by placing the duck on the red sphere and using a blue cube instead of a blue cylinder behind both objects. (Second judge pass, order swapped — scores are the average of both: Model B closely matches the prompt: the matte green cube is to the right of the glossy red sphere, the small blue cylinder is behind them, and the yellow rubber duck sits on top of the cube with correct color binding. Model A is more stylized and attractive in some ways, but it breaks key spatial and object constraints by placing the duck on the sphere, using a blue cube instead of a cylinder, and giving the green cube an incorrect shape/orientation.)
Perspective & scale
A photorealistic one-point-perspective shot down a long empty library aisle, shelves receding to a single vanishing point, books getting consistently smaller with distance and nothing warped or wrongly sized, warm overhead light, 16:9.


Model B better matches the requested one-point perspective down a long empty library aisle, with shelves and books receding more consistently to a single vanishing point and a stronger 16:9 presentation. Model A is attractive and clean, but its framing feels less convincingly long and the oversized circular ceiling lights make the perspective slightly less natural. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested one-point-perspective shot down a long empty library aisle, with shelves converging cleanly to a single vanishing point and book scale diminishing consistently without visible warping. Model A is attractive and well lit, but the aisle feels shorter, the perspective is less exact, and the square framing misses the requested 16:9 format.)
Matchup powered by OpenRouter.